Work Experience
VMware Inc., Palo Alto, USA
Member of Technical Staff, Intern, High Availability Team
Summer 2014
Project Description - I am currently working with the High Availability team on their distributed VM availability services for vSphere.
North Carolina State University
Teaching Assistant, Introduction to Artificial Intelligence(CSC 411)
Fall 2013
Description - Worked as a teaching assitant under Dr. Robert St. Amant for his undergraduate course - "Introduction to Artificial Intelligence" (CSC 411).
Indian Institute of Technology, Delhi
Research Intern working with Dr. K. K. Biswas
Summer 2012
Project Description -
A Hybrid Random Fourier Features for Large Kernel Machines - Designed and implement a modified version of Localized Linear Support Vector Machine (SVM) which is faster and more accurate than standard libraries. The modified SVM first clusters the data using K-NN and K-Means clustering algorithm and then applies a Sequential Minimal Optimization (SMO) based SVM on each cluster, thereby deriving a consensus SVM model that is later used for classification and prediction purpose. Random Fourier Features were used for feature selection and optimization. Results showed about 10-20% improvement in the classification time and the system achieved marginal improvement in classification accuracy as well, under experiments that involved running this tool against available libraries like LIBSVM and LLSVM on standard benchmark datasets. [Report]
Jaypee Institute of Information Technology, India
Teaching Assistant, Data Structures Lab
Spring 2013
Description - Worked as a teaching assitant under Dr. Manish Thakur for his undergraduate lab course - "Data Structures".
Indian Institute of Technology, Delhi
Research Intern working with Dr. K. K. Biswas
Summer 2011
Project Description -
Implementing Sequential Minimal Optimization (SMO) using a Hash Table - I designed and implemented the Sequential Minimal Optimization (SMO) algorithm for training SVM using a "hash table" and show improved results over an array implementation. Final analysis was done by running the modified SMO based SVM over standard datasets and showing improved results in terms of computation time when compared to standard libraries like LIBSVM. Results showed around 30% improvement in the classification time. [Report]